Imputation and Weighting for Survey Data

Description:

After survey data are collected and before analysis, it may be necessary to use imputation and weighting strategies.

Item non-response is best handled with imputation. The course starts by looking at the pros and cons of older simplistic methods of imputation, then more sophisticated methods and ends this section with a discussion of multiple imputation and maximum likelihood as an alternative.

Unit non-response is best handled with weighting. But weighting is also used for unequal selection probabilities, calibrating to population totals, etc. This course considers all the stages and aspects of weighting.

For both of these topics, the course presents the debates in the literature, but also includes real life examples. The course does not focus on specific software but rather on the decisions to be made in choosing an appropriate method and the principles of implementing a method. Practical workshop sessions with both discussion and some calculations will help to put learning into practice.
 
Learning objectives

By the end of the course, participants will:

• Have knowledge of a variety of data imputation methods and their pros, cons and trade-offs,
• Have knowledge of simple, sophisticated and complex weighting schemes and their pros, cons and trade-offs.
• Be able to make informed choices about which particular imputation method(s) and which particular weighting method(s) are best for their data.

Topics

• Item missing data
• Imputation strategies (simplistic methods, more sophisticated methods through to multiple imputation and the use of maximum likelihood for imputation – pros and cons of different methods)
• Are your data missing at random? (MCAR, MAR, NMAR)
• Review of basic cell weighting
• 3 stages of weighting (1 obtaining design weights; 2 compensating for non-response, 3 calibrating to known population totals)
• Examples of complex weighting schemes
• Further details on weighting: stage 2 unknown eligibility and non-response (deterministic [cell weighting] versus stochastic [response propensity score weighting]; propensity score versus classification trees), stage 3 calibration methods (GREG estimator, ratio estimator (single auxiliary variable), post-stratified estimator, raking ratio estimator).
• New developments in weights

Who will benefit?

Both participants who want to use imputation and weighting strategies on their data but also participants who want some knowledge of these topics for critiquing other’s work.

Participants need knowledge and experience of survey statistics and sampling. The SRA course on “Sampling and an Introduction to Weighting” would also be a good prerequisite.
 
Learning outcomes

Participants will:

• Have a better knowledge of post data collection adjustment strategies,
• Will also have knowledge of the pitfalls of each of the methods discussed and
• Participants will see how this information can lead to better quality survey results.

Course tutor

Dr Pamela Campanelli is a Survey Methods Consultant and runs The Survey Coach business. She is also a Chartered Statistician, Chartered Scientist and Fellow of the Academy of Social Sciences. She has worked at the University of Michigan, the U.S. Bureau of the Census, ISER at the University of Essex, and NatCen Social Research. She has both led and been a team member on ESRC grants (one on survey non-response and one on measurement error in mixed mode surveys). She regularly teaches short courses in the UK and abroad for government departments, survey research companies, universities, as well as for various other institutions and businesses.

Cost:

£260. Members of SRA receive a 25% discount and pay £195.

Website and registration:

www.the-sra.org.uk/training

Region:

Greater London

Keywords:

Quantitative Data Handling and Data Analysis, Imputation , Weighting

Related publications and presentations:

Quantitative Data Handling and Data Analysis

Back to archive...